Codex

Codex "The site-safety policy blocks this action; no user permission prompt or Auto-review was attempted."

Last checked

The error

The site-safety policy blocks this action; no user permission prompt or Auto-review was attempted.

Seen when the agent uses the browser or Computer Use on certain websites, in the Codex desktop app and ChatGPT.

OpenAI blocks the agent from acting on some websites, and this block comes before any permission prompt, so "Always allow" and Auto-review never get a say. Reports on openai/codex name Avito, Taobao, 1688, AliExpress, WeChat, Naver blog, dzen.ru, VK ID and others. OpenAI's browser docs say only that some websites block access. As of 10 November 2026 the reports are open with no OpenAI reply. Use the site's API or export, do the blocked step yourself, or use a different browser tool for that site.

Why it happens

A site-level policy in Codex's browser and Computer Use tools. When the target site is on the blocked list, the action is refused immediately, without a permission prompt or Auto-review.

  1. The site is on OpenAI's blocked list. openai/codex#59394 (opened 4 November 2026) shows avito.ru blocked through Computer Use in Chrome on Codex desktop 26.1029 and 26.1103, while kufar.by worked as a control.
  2. Marketplaces, social networks and some media sites are common in the reports. More than 70 site-safety issues were open in November 2026.
  3. Always allow does not apply. Several reports, including #59436, #59813 and #53349, show the block even with Always allow set.

The fix

  1. 1 Do not keep retrying or changing permission settings: the block is server-side and settings cannot lift it.
  2. 2 Use the site's official API or data export where it has one, and give the agent the result.
  3. 3 Split the task: you do the blocked step on the site, then hand the files or text to the agent.
  4. 4 For browsing tasks you are allowed to automate, use a different tool for that site, such as a browser automation script you run yourself.
  5. 5 Add your case to the matching issue (for Avito, #59394 and #58511) so OpenAI sees which sites matter.

Still failing?

  • Every site fails, not one. That is not this policy: check whether Computer Use is enabled and connected.
  • File edits fail with a read-only sandbox message. See that page.

Related errors

Full guideClaude Code and ChatGPT rate limit alternatives

Hit a different error?

Paste any agent error and get the cause and fix in seconds.

Open the decoder

Frequently asked questions

What does "The site-safety policy blocks this action" mean?

OpenAI does not allow the agent to act on that website, so the action was refused before any permission prompt. Your settings did not cause it and cannot override it.

Why doesn't Always allow fix it?

Always allow only answers permission prompts. This policy runs before the prompt, so nothing is asked. Several open reports confirm the block persists with Always allow.

Which sites trigger the site-safety policy?

OpenAI does not publish a list. Open reports name Avito, Taobao, 1688, AliExpress, WeChat, Naver blog, dzen.ru, music.163.com, VK ID and MyAnimeList, among others.

Is there a workaround?

Use the site's API or export, do the blocked step yourself and hand the result to the agent, or automate that site with another tool you control and are permitted to use.

Stop firefighting agent errors

Decoding errors one at a time is the manual version of what BetterClaw automates. Run your agents on a no-code AI agent platform with managed models, retries and config validation built in.

Deploy Your Agent — Start FreeBook a Demo

Free plan available · Pro $49/mo · BYOK · 7-day free trial